hidden digital watermarks in images - image processing

58 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 8, NO. 1, JANUARY 1999

Hidden Digital Watermarks in ImagesChiou-Ting Hsu and Ja-Ling Wu,Senior Member, IEEE

Abstract—In this paper, an image authentication techniqueby embedding digital “watermarks” into images is proposed.Watermarkingis a technique for labeling digital pictures by hidingsecret information into the images. Sophisticated watermarkembedding is a potential method to discourage unauthorizedcopying or attest the origin of the images. In our approach, weembed the watermarks with visually recognizable patterns intothe images by selectively modifying the middle-frequency partsof the image. Several variations of the proposed method will beaddressed. The experimental results show that the proposed tech-nique successfully survives image processing operations, imagecropping, and the Joint Photographic Experts Group (JPEG)lossy compression.

Index Terms—Digital watermark, discrete cosine transform,JPEG compression, pseudorandom permutation.

I. INTRODUCTION

DUE TO THE rapid and extensive growth of electronicpublishing industry, data can now be distributed much

faster and easier. Unfortunately, engineers still see immensetechnical challenges in discouraging unauthorized copying anddistributing of electronic documents [1]. Conventionally, apainting is signed by the artist to attest the copyright, anidentity card is stamped by the steel seal to avoid forgery, andpaper money is identified by the embossed portrait. Such kindsof handwritten signatures, seals, or watermarks have been usedsince ancient times as a way to identify the source or creator ofa document or picture. However, in the digital world, digitaltechnology for manipulating images has made it difficult todistinguish the visual truth. “Seeing is believing” will becomean anachronism [2].

One potential solution for claiming the ownership is touse electronic stamps or so-calledwatermarks, which areembedded into the images, and have the following features:

• undeletable by hackers;• perceptually invisible, i.e., the watermark should not

render visible artifact;• statistically undetectable;• resistant to lossy data compression, e.g., the Joint Photo-

graphic Experts Group (JPEG) compression;• resistant to image manipulation and processing opera-

tions, e.g., cut-and-paste, filtering, etc.

Manuscript received August 23, 1996; revised March 6, 1998. The associateeditor coordinating the review of this manuscript and approving it forpublication was Prof. Dmitry B. Goldgof.

C.-T. Hsu is with the Communication and Multimedia Laboratory, Depart-ment of Computer Science and Information Engineering, National TaiwanUniversity, Taipei, Taiwan, R.O.C. (e-mail: [email protected]).

J.-L. Wu is with the Communication and Multimedia Laboratory, Depart-ment of Computer Science and Information Engineering, National TaiwanUniversity, Taipei, Taiwan, R.O.C. (e-mail: [email protected]).

Publisher Item Identifier S 1057-7149(99)00215-8.

In the literature, several techniques have been developedfor watermarking. In [3], three coding methods for hidingelectronic marking in document were proposed. In [4]–[7],the watermarks are applied on the spatial domain. The majordisadvantage of spatial domain watermarking is that a com-mon picture cropping operation may eliminate the watermark.Other than spatial domain watermarking, frequency domainapproaches have also been proposed. In [8], a copyrightcode and its random sequence of locations for embeddingare produced, and then superimposed on the image based ona JPEG model. In [9], the spread spectrum communicationtechnique is also used in multimedia watermarking.

In most of the previous works [8]–[10], the watermark is asymbol or an random number which comprises of a sequenceof bits, and can only be “detected” by employing the “detectiontheory.” That is, during the verification phase, the original im-age is subtracted from the image in question, and the similaritybetween the difference and the specific watermark is obtained.Therefore, an experimental threshold is chosen and comparedto determine whether the image is watermarked. In this paper,we propose a technique for embedding digital watermarkswith visually recognizable patterns into the images. Since,in daily life, one claim a document, a creative work, and soon, by signing one’s signature, stamping a personal seal oran organization’s logo, such kinds of visually recognizablepatterns are more intuitive for representing one’s identity thana sequence of random numbers is. More specifically, during theverification phase of our work, an “extracted” visual pattern inconjunction with the similarity measurement will be providedfor verification.

First of all, the watermark is generated as a binary pattern,and then permuted to disperse the spatial relationship and toincrease the invisibility based on the characteristics of images.Also, since human eyes are more sensitive to lower frequencynoise, intuitively the watermark should be embedded intothe higher frequency components to achieve better perceptualinvisibility. However, since the energy of most natural imagesare concentrated on the lower frequency range, the informationhidden in the higher frequency components might be discardedafter quantization operation of lossy compression. Therefore,to invisibly embed the watermark that can survive the lossydata compression, a reasonable trade-off is to embed thewatermark into the middle-frequency range of the image.In our scheme, watermarks are embedded by modifying themiddle-frequency coefficients within each image block of theoriginal image in considering the effect of quantization. Theexperimental results show that the proposed technique couldsurvive several kinds of image processing and the JPEG lossycompression.

1057–7149/99$10.00 1999 IEEE

HSU AND WU: DIGITAL WATERMARKS 59

This paper is organized as follows. The embedding approachis described in Section II. Section III describes the watermarkextraction method. In Section IV, the experimental results areshown. In Section V, several issues of the proposed method arediscussed. The conclusion of this paper is stated in Section VI.

II. EMBEDDING METHODS

In our approach, a block DCT-based algorithm is developedto embed the image watermarking.

Let be the original gray-level image of size , andthe digital watermark be a binary image of size . Inthe watermark, the marked pixels are valued as one’s, and theothers are zero’s. Since only the middle-frequency range of thehost image will be processed during the watermark embedding,the resolution of a watermark image is assumed to besmaller than that of the original image. For example, foreach 8 8 image block, only (64 ) coefficients willbe used for the watermark embedding. The ratio ofand determines the amount of information tobe embedded into the image. In general, for more robustand invisible embedding, the amount of information can beembedded should be reduced. On the other hand, in orderto provide a visually recognizable watermark with nontrivialamount of information, instead of using an ID number withtrivial amount of data, making the watermark embeddingperceptually invisible is not a trivial problem.

The original image and digital watermark are repre-sented as

(1)

where is the intensity of pixeland is the number of bits used in each pixel.

(2)

where .In , there are image blocks with size 8 8. To

obtain the same number of blocks as the image ,the watermark is decomposed into several blocks with size

. For example, if and, the block size of the watermark block is 4 4,

and if and , the block size of thewatermark block is 2 2. The extra columns and rows mightbe added to complete each image and watermark blocks.

A. Pseudorandom Permutation of the Watermark

In the approach, each watermark block is embedded intothe middle-frequency range of each image block using block-transform instead of full-frame transform. Therefore, eachwatermark block will only be dispersed over its correspondingimage block, instead of the entire spatial image. Obviously,without appropriate adjustment for the spatial relationship ofthe watermark, a common picture-cropping operation mayeliminate the watermark.

To survive picture-cropping, a fast two-dimensional (2-D)pseudorandom number traversing method is used to permute

the watermark to disperse its spatial relationship, i.e.,

Permute

and (3)

where pixel is permuted to pixel in a pseudoran-dom order.

In our approach, the permutation is implemented as follows.First, number each pixel from zero to . Second,generate each number in random order. Finally, generate thecoordinate pairs by mapping the random sequence numberinto a 2-D sequence. For example, for a digital watermarkof size 128 128, use a “linear feedback shift register”[11] to generate a random sequence from 1 to 16 383. Then,for each sequence element number, computeand as the permuted vertical and horizontalcoordinates.

B. Block-Based Image-Dependent Permutationof the Watermark

In order to improve the perceptual invisibility, the charac-teristics of the original image should be considered, e.g., themodifications of high frequencies or high luminance regionsare less perceptible. Such image-dependent properties can beused to shuffle the pseudorandom permuted watermark to fitthe sensitivity of human eyes.

For each image block of size 8 8, the variances (which isused as a measure of invisibility under watermark embedding)are computed and sorted. For each watermark block of size,

, the amount of information (i.e.,the number of signed pixels) are sorted also. Then, shuffleeach watermark block into the spatial position according thecorresponding sorting order of the image block, i.e.,Permute ; in which

(4)

and

(5)

where block is shuffled to block by the block-based permutation. Fig. 1 shows an example of the sortingand the permutation.


Fig. 1. Example of block-based image-dependent permutation, where foreach 8� 8 image block, the variance is computed and sorted, and for eachwatermark block of size(M1�

8

N)�(M2�

8

N), the amount of information

(i.e., the number of signed pixels) is sorted also, and then each watermarkblock is shuffled into the spatial position according the corresponding sortingorder of the image.

C. Block Transformation of the Image

Since the discrete cosine transform (DCT) used by JPEG[12] is performed on blocks of 8 8, the input imageis divided into blocks of 8 8, and each block is DCTtransformed independently. That is,

where FDCT denotes the operation of forward DCT.

D. Choice of Middle-Frequency Coefficients

The human eye is more sensitive to noise in lower frequencycomponents than in higher frequency ones. However, theenergy of most natural images are concentrated in the lowerfrequency range, and the information hidden in the higherfrequency components might be discarded after quantizationoperation of lossy compression. In order to invisibly embedthe watermark that can survive lossy data compressions, areasonable trade-off is to embed the watermark into themiddle-frequency range of the image. To this end, for each8 8 image block, only (64 ) coefficients are selectedout of the 64 DCT coefficients. Those selected coefficients arethen mapped into a reduced image block of size

. That is, the middle-frequency coefficients selectedfrom the image of size are collected to composea reduced image of size , which has the sameresolution with the binary watermark.

Reduce

where

(6)

and

(7)

(a) (b)

Fig. 2. Example of defining the middle-frequency coefficients, in which thecoefficients are picked up in zigzag-scan order and then reordered into blockof 4 � 4. (a) Zigzag ordering of DCT coefficients and the middle frequencycoefficients are shown in the shadow area. (b) Picked up coefficients aremapped into the 4� 4 block.

Fig. 3. Residual mask, where each square includes a reduced image blockof size (M1 �

8

N) � (M2 �

8

N), and positione stands for the current

reduced block.

(a) (b)

Fig. 4. Luminance quantization table. (a) Default JPEG quantization table.(b) JPEG quantization table used by Image Alchemy, Handmade Software Inc.

For example, if and , only 16 DCTcoefficients are processed during the watermark embedding,and the other 48 DCT coefficients are left unchanged. Fig. 2exemplifies our definition of the middle-frequency coefficients,which are mapped into a reduced block of size 44.

E. Modification of the DCT Coefficients

Now, a permuted digital watermark and a reduced image(which contains only the middle-frequency components of theoriginal image) both with size are obtained. Foreach watermark block of size , thereduced image block of size at thecorresponding spatial position will be modified adequately toembed the watermarked pixels.

In our opinion, embedding each watermarked pixel bymodifying the polarity between the corresponding pixels inneighboring blocks is an effective approach to achieve theinvisibility and survival for low compression ratio of JPEG.However, this method is not robust with respect to the com-pression attacks with higher compression ratio . At part(a), we will address the technical challenges. At part (b), an


Fig. 5. Watermark embedding steps.

improved method that is resistant to higher compression ratiowill be described.

1) Embedding into the Relationship Between NeighboringBlocks: A 2-D residual mask is used to compute the polarityof the chosen middle-frequency coefficients between neigh-boring blocks. For example, in Fig. 3, if

, then the polarity is a binary pattern (zeroor one) which represents the coefficients at the position ofthe current reduced-block is largerpolarity or lesspolarity than the coefficient at the corresponding

position of the previous reduced-block. That is,

Polarity

(8)


Fig. 6. Watermark extracting steps.

where

if

otherwise.(9)

After the binary polarity pattern is obtained, for each markedpixel of the permuted watermark, modify the DCT coefficientsaccording the residual mask to reverse the corresponding

polarity. That is,

and(10)

where

ifif

(11)

Then, construct from such that the differences betweenand are minimized or smaller than a user specified


threshold;

Expand

s.t. threshold (12)

Note that the “Expand” operation constructs based on po-larity . For example, assign the initial coefficient

, and add (or subtract) the coefficients of neighboringblocks according to the residual mask in order to match thecorresponding polarity . Then, proceed to successivecoefficients by modifying only those who will not affect thepolarities of the previous-processed coefficients.

In order to improve the invisibility, the polarity should becomputed for absolute value of the coefficients so that thesign (plus and minus signs) of the coefficients are hopefullypreserved to reduce the changes introduced by modification.

Besides, in order to survive the JPEG lossy compression, thequantization effect utilized in the JPEG codec must be con-sidered. Fig. 4(a) shows the suggested luminance quantizationtable for JPEG standard, which usually cause perceptible ar-tifacts when viewed on high-quality displays. Fig. 4(b) showsanother quantization table used in most JPEG software. Thevalues are almost half of the corresponding JPEG suggestedquantization values. Based on a referenced quantization table,the polarity are computed from coefficients after quantizationand then dequantization. Therefore, in case of quantizationattack, the correct marked pixel should be extracted. That is,the polarity should be

if

otherwise(13)

where is the quantization value at the correspondingposition .

However, since quantization tends to make many coeffi-cients zero (especially those for higher spatial frequencies),if the modification of the coefficients is not large enough,most middle-frequency coefficients will also be truncated tozero after coarse quantization. Besides, in order to preservethe polarity to survive quantization according the specifiedresidual mask, not only those middle-frequency coefficientsin the current block have to be modified, but also all theneighboring blocks involved in the residual mask have be tomodified with the same volume. Therefore, although survivalfor lossy compression, the watermarked image which areembedded with large modification will not be perceptuallyequivalent to the original image.

2) Embedding into the Relationship Within Each BlockInorder to overcome the technical challenges addressed above,while still hopefully not to propagate the modifications intothe neighboring blocks (so as to improve the invisibility), the

(a) (b)

(c) (d)

Fig. 7. Example of the proposed watermarking approach. (a) Test imageLena. (b) Watermark. (c) Watermarked image (with PSNR= 40.83 dB). (d)Extracted watermark (withNC �= 1).

relatively more reliable DC coefficient (instead of the middle-frequency coefficients of neighboring blocks) is used as areference value for each block. That is,

Polarity

(14)

where

if

otherwise.(15)

For each marked pixel, add (or subtract) the correspondingcoefficient so that the modified coefficients will have thereverse polarity .

F. Inverse Block Transform

Finally, map the modified middle-frequency coefficientsinto to get . Then, inverse DCT (IDCT) of the associated


(a) (b)

Fig. 8. Another example of the proposed approach. (a) Cameraman test image. (b) Watermarked image (with PSNR= 39.29 dB).

result to obtained the embedded image.

Fig. 5 illustrates the steps of embedding the watermark, whereLena is used as the test image of size 256256, and aseal with Chiou-Ting Hsu’s Chinese name is used as a binarywatermark of size 128 128.

III. W ATERMARK EXTRACTING METHOD

The extraction of watermark requires the original image,the watermarked image, and either the watermark or thepermutation mapping used in image-dependent permutationduring the embedding steps. The extraction steps are describedas follows.

A. Block Transformation

Both the original image and the image in question areDCT transformed.

B. Generation of Polarity Patterns

Generate the reduced images which contain only the middle-frequency coefficients and then use these middle-frequencyDCT coefficients to produce the polarity patterns. That is,

Reduce

Reduce

and then

Polarity

Polarity

(a) (b)

Fig. 9. (a) Blurred version of Fig. 7(c). (b) Extracted watermark withNC = 0:982.

C. Extract the Permuted Data

Perform exclusive-or (XOR) operation on these two polaritypatterns to obtain a permuted binary data, i.e.,

where

(16)

D. Reverse Block-Based Image-Dependent Permutation

The image-dependent permutation mapping could be ob-tained either by saving as a file during the embedding steps orrecomputed from the original image and the watermark. Basedon the mapping, reverse permute to get .

E. Reverse Pseudorandom Permutation

Reverse-permute to get the watermark

(17)

where is reverse-permuted to according to thepredefined pseudorandom order.


(a) (b)

Fig. 10. (a) Image enhanced version of Fig. 7(c) with slightly enhancedcontrast. (b) Extracted watermark withNC = 0:9985.

(a) (b)

Fig. 11. (a) Image enhanced version of Fig. 7(c) with strongly enhancedcontrast. (b) Extracted watermark withNC = 0:973.

F. Similarity Measurement

In our scheme, the extracted watermark is a visually recog-nizable pattern. The viewer can compare the results with thereferenced watermark subjectively. However, the subjectivemeasurement is dependent on factors such as the expertiseof the viewers, the experimental conditions, etc. Therefore,a quantitative measurement is needed to provide objectivejudgment of the extracting fidelity. We define the similar-ity measurement between the referenced watermarkandextracted watermark as

Normalized Correlation

(18)which is the cross-correlation normalized by the referencewatermark energy to give unity as the peak correlation.

Fig. 6 illustrates the procedures for extracting the water-marks.

IV. EXPERIMENTAL RESULTS

Fig. 7 shows an example of embedding and extractingresults, where Lena is used as the test image again, and apattern with “NTU CSIE CMLAB” is used as the watermark.Fig. 8 is another example, where the cameraman image iswatermarked with Fig. 7(b).

(a)

(b) (c) (d)

Fig. 12. (a) Quarter of an embedded image is discarded. And (b)–(d) are theextracted watermarks, where (b) is without pseudorandom permutation norblock-based image-dependent permutation(NC = 0:7922), (c) is with onlyblock based permutation(NC = 0:9372), and (d) is with both permutations(NC = 0:7623).

Fig. 13. Relationship between the cropping ratios and the NC values.

A. Image Processing Operation

Smoothing operations are used to diminish spurious effectswhich may be present in images from a poor transmissionchannel. Fig. 9 shows a blurred version of the watermarkedimage, and the extracted watermark, though interfered by noiseis still recognizable.

The contrast of an image is usually adjusted to enhance thesubjective quality. Figs. 10 and 11 show the results of applyingslightly and strongly enhanced operations to a watermarkedimage accordingly. The extracted results are still highly similarto the original watermark.

B. Image Cropping Operation

During the image manipulation, the uninterested part ofan image is usually cropped. As described in Section II-A, apseudorandom permutation is performed to disperse the spatialrelationship of the watermark. Therefore, it would be hard for


(a) (b) (c) (d)

Fig. 14. Extracted watermarks of the cropped versions of Fig. 7(c). (a) A quarter of the image is cropped, and the missing portions is filled with zerovalues (NC = 0:7623). (b) A quarter of the image is cropped, and the missing portion is filled with unwatermarked image(NC = 0:7339). (c)Half of the image is cropped, and the missing portions is filled with zero values(NC = 0:5964). (d) Half of the image is cropped, and the missingportions is filled with unwatermarked image(NC = 0:5276).

(a) (b) (c) (d)

Fig. 15. Extracted watermarks of JPEG compressed version of Fig. 7(c), where (a) with compression ratio 5.92 andNC = 0:99, (b) with compression ratio7.16 andNC = 0:883, (c) with compression ratio 8.46 andNC = 0:726, and (d) with compression ratio 9.05 andNC = 0:661.

TABLE ICHANGES OF NC VALUES UNDER JPEG LOSSYCOMPRESSION. AS THE COMPRESSIONRATIO INCREASES FROM3.49 TO 10.74,THE CORRESPONDINGIMAGE

QUALITY IS DROPPED FROM33.78 dBTO 31.27 dB,AND THEREFORE, THE NC VALUES DECREASE FROM0.999TO 0.413, ACCORDINGLY.

a “pirate” to detect or remove the watermark by cutting somepart of the image. For example, in Fig. 12(a), a quarter ofthe watermarked image is discarded. If neither pseudorandomnor block-based image dependent permutation was appliedduring the embedding, the extracted watermark [as shown inFig. 12(b)] will reveal the spatial information of the watermarkembedding. However, once the permutation is introduced, thelost information will be distributed over the whole image, andthe extracted error will also be distributed over the wholeresult. In such case, the results will be interfered with noisesbut will not reveal the spatial position of the watermark [asshown in Fig. 12(c) and (d)] Note that, since Fig. 12(c) iswithout using the pseudorandom permutation, the noises areless random then 12(d) does.

Fig. 13 shows the relationship between the similarity mea-surement NC and the cropping ratio. Since a pseudorandompermutation is applied, the effect of cropping operation to theNC of the extracted results is almost linear.

Fig. 14 shows the extracted results from various croppedimage. Fig. 14(a) and (c) are extracted from cropped imageswhere the missing portions are filled with zero values, andFig. 14(b) and (d) are extracted from cropped images wherethe missing portions are filled with original unwatermarkedimages. As shown in the figure, filling with zero values in themissing portions distributes more noises over the entire resultsand influences the visual recognition.

Fig. 16. If M1 = N1=4 and M2 = N2=4, then there are only fourmiddle-frequency coefficients should be chosen for each watermark. Thesethree watermarks would be embedded into one image with their individualuser key, or identical watermark could be repeatedly embedded three timesto improve the robustness.

C. JPEG Lossy Compression

Fig. 15 shows the extracted results from JPEG compressedversion of the watermarked images with compression ratio5.92, 7.16, 8.46, and 9.05. The quantization table in Fig. 4(a)is used as the referenced quantization values as describedin Section II-E. Table I shows that as the compression ratioincreases, the NC value decrease accordingly. Therefore, asthe compression ratio are high enough to quantize DCTcoefficients very coarsely, the watermark will be destroyed and


(a) (b) (c)

(d) (e)

Fig. 17. Example of multiple watermarking using three watermarks with size 128� 128 as shown in (a), (b), and (c) as watermarks 1–3 of Fig. 16accordingly, where (d) is the host image with size of 512� 512, and (e) is the multiple-watermarked image with PSNR= 39.48 dB.

become indiscernible. However, in this situation, the qualityof the JPEG compressed image (without being watermarked)will be degraded severely so that the processes of digitalwatermarking become less meaningful.

V. DISCUSSION

A block DCT-based watermarking technique for images isproposed in this work. There are some issues are worthy ofgiving further discussion as follows.

A. Image Dependent Permutation

In Section II-B, a block-based image-dependent permuta-tion according to the characteristics of both the image andwatermark is used in our work. Although the watermark isembedded into the middle-frequency coefficients, for thoseblocks with little variances (i.e., the blocks containing lowfrequency contents), the modification of DCT coefficients willintroduce visible artifacts. To reduce the artifacts, we sortthe variances of image blocks and the amount of informationwithin each watermark block, and then map the watermarkblocks with more signed pixels into those image blocks withhigher variances.

However, based on our simulation, no gain of objectivequality (such as PSNR) can be obtained by using these kindsof permutations. However, better subjective quality, especiallyfor those smoother parts of the image, could be obtained.

B. User Key

A “user key” is provided as a secret key that can beused to serve various embedding processes by using the

same embedding technology. Besides, a user key is treatedas a parameter during the extracting steps. In the proposedembedding method, a “user key” should defines the followingissues.

1) Seed of the pseudorandom number generator:The seed defines the initial position of an pseudorandompermutation, which could be any one between one and

.2) Choice of middle-frequency coefficients:

There are middle-frequency coefficientsshould be picked out in all of 64 DCT coefficientsfor each block. In the user key, the coefficients to beprocessed should be specified.

3) Mapping of the chosen coefficients into a reduced block:Fig. 2(b) shows one way to map the chosen coefficients.Other kinds of mapping are also possible.

C. Size of Watermarks

With visually recognizable watermarks, there are nontrivialamount of information will be embedded into the originalimage. Obviously, once the ratio of andbecomes larger, the number of DCT coefficients involvedin the embedding will be smaller. In such case, the invisi-bility will be improved. In addition, multiple watermarkingor repeatedly embedding identical watermarks to harden therobustness is also possible. For example, ifand , then there are only four middle-frequencycoefficients should be chosen from 64 coefficients for thewatermark. As shown in Fig. 16, three watermarks could beembedded into an image with different user key which choosesdifferent coefficients from the image block. Fig. 17 shows an


(a)

(b)

Fig. 18. Extracted watermarks from JPEG compressed version of Fig. 17(e)with individual user keys, where (a) with compression ratio 5.58 and NCvalues are 0.996, 0.993 and 0.975 accordingly, and (b) with compressionratio 7.13 and NC values are 0.995, 0.983, and 0.892 accordingly.

example of multiple watermarked result from Fig. 16, andFig. 18 shows the extracted results from JPEG compressedimage of Fig. 17(e) with different user key.

VI. CONCLUSION

This paper has presented a technique for embedding digitalwatermark into the images. The embedding and extractingmethods of the DCT-based approach have been described.

The experimental results show the proposed embeddingtechnique can survive the cropping of an image, image en-hancement and the JPEG lossy compression. By carefullydefining the “user key,” multiple watermarking and repeat-edly embedding to harden the robustness are available. Ourtechnique could also be applied to the multiresolution imagestructures with some modification about the choice of middle-frequency coefficients [13].

Other kinds of attacks, such as image resampling and imagerotation, are still challenging to our current work, and havebeen chosen to be the major direction of our future work.

REFERENCES

[1] B. M. Macq and J. J. Quisquater, “Cryptology for digital TV broadcast-ing,” Proc. IEEE, vol. 83, pp. 954–957, June 1995.

[2] W. J. Mitchell, “When is seeing believing?,”Sci. Amer., pp. 68–73,Feb. 1994.

[3] J. T. Brassil, S. Low, N. F. Maxemchuk, and L. O’Gorman, “Electronicmarking and identification techniques to discourage document copying,”IEEE J. Select. Areas Commum., vol. 13, pp. 1495–1504, Oct. 1995.

[4] I. Pitas and T. H. Kaskalis, “Applying signatures on digital images,”in Proc. IEEE Nonlinear Signal and Image Processing, June 1995, pp.460–463.

[5] O. Bruyndonckx, J. J. Quisquater, and B. Macq, “Spatial method forcopyright labeling of digital images,” inProc. IEEE Nonlinear Signaland Image Processing, June 1995, pp. 456–459.

[6] S. Walton, “Image authentication for a slippery new age,”Dr. Dobb‘sJ., pp. 18–26, Apr. 1995.

[7] W. Bender, D. Gruhl, and N. Morimoto, “Techniques for data hiding,”Proc. SPIE, vol. 2420, p. 40, Feb. 1995.

[8] E. Koch and J. Zhao, “Toward robust and hidden image copyrightlabeling,” in Proc. IEEE Nonlinear Signal and Image Processing, June1995, pp. 452–455.

[9] I. J. Cox, J. Kilian, T. Leighton, and T. Shammoon, “Secure spreadspectrum watermarking for multimedia,” Tech. Rep. 95-10, NEC Res.Inst., Princeton, NJ, 1995.

[10] M. D. Swanson, B. Zhu, and A. H. Tewfik, “Transparent robust imagewatermarking,” inProc. ICIP’96, pp. 211–214.

[11] B. Sklar, Digital Communications. Englewood Cliffs, NJ: Prentice-Hall, 1988.

[12] W. B. Pennebaker and J. L. Mitchell,JPEG: Still Image Data Com-pression Standard. New York: Van Nostrand Reinhold, pp. 34–38,1993.

[13] C.-T Hsu and J.-L. Wu, “Multiresolution watermarking for digitalimages,” IEEE Trans. Circuits Syst. II, vol. 45, pp. 1097–1101, Aug.1998.

Chiou-Ting Hsu received the B.S. degree incomputer and information science from NationalChiao Tung University, Hsin-Chu, Taiwan, R.O.C.,in 1991, and the Ph.D. degree in computer scienceand information engineering from National TaiwanUniversity (NTU), Taipei, Taiwan, in 1997.

She is currently a Post-doctoral Researcher withthe Communication and Multimedia Laboratory,Department of Computer Science and InformationEngineering, NTU. Her research interests arein digital watermarking, multiresolution signal

processing, and image/video coding.

Ja-Ling Wu (S’85–A’87–SM’98) was born inTaipei, Taiwan, on November 24, 1956. He receivedthe B.S. degree in electronics engineering fromTam-Kang University, Tam-Shoei, Taiwan, R.O.C.,in 1979, and the M.S. and Ph.D. degrees in electricalengineering from Tatung Institute of Technology,Taipei, in 1981 and 1986, respectively.

From August 1986 to July 1987, he was anAssociate Professor in the Department of ElectricalEngineering, Tatung Institute of Technology. Hebecame an Associate Professor at the Department of

Computer Science and Information Engineering, National Taiwan University,Taipei, in August 1987, and a Professor in August 1990. Since August 1996,he has been with National Chi-Nan University, Puli, Taiwan, as the Chairmanof the Department of Information Engineering. He currently teaches coursesin digital signal processing and multimedia data compression and conductsresearch in the areas of signal processing, image/video coding, and digitalwatermarking techniques. He has authored more than 150 technical papersin these areas.

Dr. Wu received the Outstanding Research Award of the National ScienceCouncil of the Republic of China from 1987 to 1994, the OutstandingYouth Medal of the Republic of China in 1989, the Award for 1993 R.O.C.Distinguished Information People of the Year, the Special Long-Term Awardfor Collaboratory Research, sponsored by the Acer Corporation in 1994,the Best Paper Award for the R.O.C. Association of Image Processingand Multimedia Applications in 1995, and the Long-Term Medal for TenDistinguished Researchers in 1996.

hidden digital watermarks in images - image processing

Documents